Thursday, February 28th 2013

AMD Working On Stripped-Down PlayStation 4 SoC for PCs

Ahead of its unveiling last week, it was expected that Sony's PlayStation 4 console would be driven by little more than an AMD A-Series "Trinity" APU. It ended up being a lot more than that. The custom-design SoC that drives the next-generation console is a joint effort between AMD and Sony, which integrates an 8-core x86-64 CPU based on the company's new "Jaguar" micro-architecture; a GPU based on its Graphics CoreNext technology; a GDDR5 integrated memory controller, and certain enhancements by Sony. In an interview with The Inquirer, the company hinted that it's interested in porting the SoC over to the PC platform, minus Sony's share of the development.
PlayStation 4, although based on the x86 CPU machine-architecture, doesn't conform to any known PC specification. It uses 8 GB of GDDR5 memory as both system and graphics memory, several of its interfaces are out of specs of anything that can be implemented on a PC motherboard. Therefore, its SoC can't simply be soldered onto a PC motherboard. AMD would have to first strip the SoC of Sony's share of the development (or risk having to license it).

Next up, it would have to strip the chip of its most interesting component, the unified GDDR5 IMC. JEDEC does not have a GDDR5 DIMM specification, nor would motherboard makers be interested in hard-wiring expensive GDDR5 memory chips on to their motherboards and render the memory subsystem non-expandable. The PlayStation 4 SoC uses a 256-bit wide GDDR5 memory interface, with a stellar memory bandwidth of 176 GB/s. That's over six times the memory bandwidth of an Intel Core i7 "Ivy Bridge" machine running dual-channel DDR3-1600 MHz. A fallback to DDR3 could hence greatly cripple the SoC. It would be extremely interesting to see how AMD handles the checks-and-balances needed to bring the SoC over to the PC.Sources: The Inquirer, The TechReport
Add your own comment

70 Comments on AMD Working On Stripped-Down PlayStation 4 SoC for PCs

#1
mastrdrver
I never said they were the same and neither did AMD. Though, you'll never be able to get a Jaguar APU with 8 cores and the GPU power of a 7850. AMD said that a "stripped down version" is coming. Of course there is, and for anyone that has seen AMD public road maps, this no surprise.

Kabini and Temash uses the same "Jaguar cores" (per an AMD slide) that the PS4 chip uses.

High end Kabini will be a little under 2Ghz. The PS4 chip is going to have a similar clock since it is based on the same architecture.

According to some leaks, Temash has similar performance to that of the current Intel i3 2367M.
Posted on Reply
#2
theoneandonlymrk
mastrdrver said:
I never said they were the same and neither did AMD. Though, you'll never be able to get a Jaguar APU with 8 cores and the GPU power of a 7850. AMD said that a "stripped down version" is coming. Of course there is, and for anyone that has seen AMD public road maps, this no surprise.

Kabini and Temash uses the same "Jaguar cores" (per an AMD slide) that the PS4 chip uses.

High end Kabini will be a little under 2Ghz. The PS4 chip is going to have a similar clock since it is based on the same architecture.

According to some leaks, Temash has similar performance to that of the current Intel i3 2367M.
Im expecting amd powerstates and boost on it myself prob 1.6ghz stck boosting to 2.6per core and within tdp
Posted on Reply
#3
Steven B
motherboard makers can hard embed GDDR5, i mean just like on a GPU's PCB. This SOC is basically a GPU with a built in CPU. I think people forget those guys who design GPUs also design CPU's and their boards. besides i doubt any real effort will be made to make that AMD thing into a mainstream product, why would there? We really need more AMD CPUs?
Posted on Reply
#4
mastrdrver
theoneandonlymrk said:
Im expecting amd powerstates and boost on it myself prob 1.6ghz stck boosting to 2.6per core and within tdp
2.6Ghz is not going to happen. The architecture is not built for speed but power.
Posted on Reply
#5
cdawall
where the hell are my stars
mastrdrver said:
2.6Ghz is not going to happen. The architecture is not built for speed but power.
It might exceed the TDP, but I do not see it having any issues exceeding 2.6ghz. It still uses off the shelf silcone.
Posted on Reply
#6
mastrdrver
cdawall said:
It might exceed the TDP, but I do not see it having any issues exceeding 2.6ghz. It still uses off the shelf silcone.
It's not a problem of TDP, it's not a problem of power, it's a problem of the architecture.

The architecture is not made for high clocks, unlike Bulldozer. It does not have enough gates to be able to clock that high.

AMD has already stated that the clocks for Jaguar will be 10% over Bobcat's normal/boost clocks and that is for the 25w Kabini part.
Posted on Reply
#7
theeldest
xenocea

Dang kids coming into my forums and causing a ruckus. You keep your music turned down, ya hear‽
Posted on Reply
#8
cdawall
where the hell are my stars
mastrdrver said:
It's not a problem of TDP, it's not a problem of power, it's a problem of the architecture.

The architecture is not made for high clocks, unlike Bulldozer. It does not have enough gates to be able to clock that high.

AMD has already stated that the clocks for Jaguar will be 10% over Bobcat's normal/boost clocks and that is for the 25w Kabini part.
The standard ARM chip is knocking on 2ghz door with quads capable of 1.6ghz. I would bet money certain Jaguar chips will be capable of hitting 2.6ghz at the top end.
Posted on Reply
#9
mastrdrver
Bobcat is not a speed monster like Bulldozer and P4 were designed to be. Thus it does not have the long gates required for high clock speeds.

As to what ARM has to do with an x86 processor, idk. Might as well compare the 1Ghz clock speed on a GK110 and say that nVidia doesn't know how to use power efficiently.
Posted on Reply
#10
cdawall
where the hell are my stars
mastrdrver said:
Bobcat is not a speed monster like Bulldozer and P4 were designed to be. Thus it does not have the long gates required for high clock speeds.

As to what ARM has to do with an x86 processor, idk. Might as well compare the 1Ghz clock speed on a GK110 and say that nVidia doesn't know how to use power efficiently.
What are you referring to with gates? CPU's have ALU's and a processing pipeline. Bulldozer is a short/mid length pipeline like all other AMD/Intel CPU's outside of P4. Pipeline length doesn't have anything to do with clockspeed. IPC and many other design cues drop into the speed side of things. :laugh:
Posted on Reply
#11
theoneandonlymrk
mastrdrver said:
Bobcat is not a speed monster like Bulldozer and P4 were designed to be. Thus it does not have the long gates required for high clock speeds.

As to what ARM has to do with an x86 processor, idk. Might as well compare the 1Ghz clock speed on a GK110 and say that nVidia doesn't know how to use power efficiently.
Arms relevance increases daily and 2.6 on a low power cpu using hurry to sleep and intelligent power gateing is a thing of the present, tried and improved on yearly one to four cores will boost to 2.6 imho but it Is just an opinion
Oh and winrt uses arm whereas Amd s Fx line emulates x86 no special hardware odd eh
Posted on Reply
#12
mastrdrver
cdawall said:
What are you referring to with gates? CPU's have ALU's and a processing pipeline. Bulldozer is a short/mid length pipeline like all other AMD/Intel CPU's outside of P4. Pipeline length doesn't have anything to do with clockspeed. IPC and many other design cues drop into the speed side of things. :laugh:
:laugh: Oh yea, I meant pipeline length. I was thinking about gate delays.

Anyway, I've always been under the understanding that clock speed was related to pipeline length since clock speed is nothing more then the calculations per second (or some calculation like that) that the processor can do. To go higher (thus higher clock speeds) you need a long pipeline length which was why P4 for so long. Problem is, long pipelines also need more power and suffer more from cache misses.

In any case, AMD has already stated that the max turbo clock for Kabini in a PC will be just under 2Ghz (10%+ on top of what Bobcat already has). Adding more cores just ups the power usage so I can't see how they would reach 2.6Ghz.

Also I've read that consoles are limited to a max power usage of something less then 50w. How true this is, I don't know.

theoneandonlymrk said:
Arms relevance increases daily and 2.6 on a low power cpu using hurry to sleep and intelligent power gateing is a thing of the present, tried and improved on yearly one to four cores will boost to 2.6 imho but it Is just an opinion
Oh and winrt uses arm whereas Amd s Fx line emulates x86 no special hardware odd eh
What does the clock speed of an ARM processor have to do with the clock speed of the AMD chip inside the PS4?
Posted on Reply
#13
cdawall
where the hell are my stars
mastrdrver said:
:laugh: Oh yea, I meant pipeline length. I was thinking about gate delays.

Anyway, I've always been under the understanding that clock speed was related to pipeline length since clock speed is nothing more then the calculations per second (or some calculation like that) that the processor can do. To go higher (thus higher clock speeds) you need a long pipeline length which was why P4 for so long. Problem is, long pipelines also need more power and suffer more from cache misses.

In any case, AMD has already stated that the max turbo clock for Kabini in a PC will be just under 2Ghz (10%+ on top of what Bobcat already has). Adding more cores just ups the power usage so I can't see how they would reach 2.6Ghz.

Also I've read that consoles are limited to a max power usage of something less then 50w. How true this is, I don't know.
The architecture itself has more to do with clockspeed than pipeline speed. Take Phenom I and Phenom II for example same basic CPU on a more mature design huge clockspeed difference. Athlon K8's represent the same thing 1ghz all the way up to 3.2ghz is a huge difference.

Kabini is a bulldozer designed CPU and a rather mature one at that it has the ability to scale to 4ghz if the TDP and bulk yields allow it.
Posted on Reply
#14
cadaveca
My name is Dave
cdawall said:
one at that it has the ability to scale to 5ghz if the TDP and bulk yields allow it.
Fix'd.


Really. BD core design made for 5 GHz. Like the song goes...You ain't seen nothing yet...b-b-b-b-b-aby...



or so I am hoping.:p
Posted on Reply
#15
NeoXF
Stripped-down PS4 SoC... sounds to me like the embedded Kaveri w/ GDDR5...


cadaveca said:

Really. BD core design made for 5 GHz. Like the song goes...You ain't seen nothing yet...b-b-b-b-b-aby...



or so I am hoping.:p
I wouldn't be surprised if 28nm Steamroller clocks at 5GHz out of the factory, considering Piledriver refresh (that's still 32nm) will prolly hit 4,5GHz or more... Still, I'd rather they clock it lower and fit in a lower TDP envelope and/or make them in 5M/10C or 6M/12C versions as well.


Also, why hasn't anyone posted about the desktop Richland moving forward to the 12th of this month? Or the specifications of the normal power mobile parts, or the fact that mobile flagship will suport DDR3-1866.

Sorry, I just happen to be very up-to-date with AMD APUs at the moment. :P
And it's kinda frustrating seeing all these crappy news articles about humdrum TITAN brand launches or misc hardware that almost no one is interested it, but nothing on Richland, which is so close to launch.
Posted on Reply
#16
theoneandonlymrk
mastrdrver said:
:laugh: Oh yea, I meant pipeline length. I was thinking about gate delays.

Anyway, I've always been under the understanding that clock speed was related to pipeline length since clock speed is nothing more then the calculations per second (or some calculation like that) that the processor can do. To go higher (thus higher clock speeds) you need a long pipeline length which was why P4 for so long. Problem is, long pipelines also need more power and suffer more from cache misses.

In any case, AMD has already stated that the max turbo clock for Kabini in a PC will be just under 2Ghz (10%+ on top of what Bobcat already has). Adding more cores just ups the power usage so I can't see how they would reach 2.6Ghz.

Also I've read that consoles are limited to a max power usage of something less then 50w. How true this is, I don't know.



What does the clock speed of an ARM processor have to do with the clock speed of the AMD chip inside the PS4?
You indicated arm was irelivant I was pointing out x86 is largely unimportant now and all these processors are made in the same way's so clock speeds dont vary that much on diff npdes and processes , the rich land apu at 2ghz is a lower bin obv but thats not to say they couldn't have been made quicker, Amd have a lot of sku's lingering in the lower price sector and have to be carefull not to diminish old stock saleability .
Ps4 will be top bin richland all 8 cores and features working and 2.6ghz it Will be.
The more I think about it though I too wouldn't be surprised by higher clocks. Efin phone
Posted on Reply
#17
TheLaughingMan
Why would you bee surprised by clocks higher than 2.6 GHz. Both current generation offerings from Microsoft and Sony are clocked at 3.2 GHz. While more cores will be a great boon since they can better utilize them, raw clock speed can be very helpful in single threaded applications such as Blu-ray playback, web browsing, system UI, their audio cloud player thingy, etc.

I personall don't see the cores being less than 3.6 GHz with Turbo up to 4.0 GHz easy.
Posted on Reply
#18
mastrdrver
theoneandonlymrk said:
You indicated arm was irelivant I was pointing out x86 is largely unimportant now and all these processors are made in the same way's so clock speeds dont vary that much on diff npdes and processes , the rich land apu at 2ghz is a lower bin obv but thats not to say they couldn't have been made quicker, Amd have a lot of sku's lingering in the lower price sector and have to be carefull not to diminish old stock saleability .
Ps4 will be top bin richland all 8 cores and features working and 2.6ghz it Will be.
The more I think about it though I too wouldn't be surprised by higher clocks. Efin phone
PS4 is not Richland. Richland does not have Jaguar cores. Jaguar is an updated Bobcat core, not Bulldozer.

TheLaughingMan said:
Why would you bee surprised by clocks higher than 2.6 GHz. Both current generation offerings from Microsoft and Sony are clocked at 3.2 GHz. While more cores will be a great boon since they can better utilize them, raw clock speed can be very helpful in single threaded applications such as Blu-ray playback, web browsing, system UI, their audio cloud player thingy, etc.

I personall don't see the cores being less than 3.6 GHz with Turbo up to 4.0 GHz easy.
AMD has already stated what max clocks will be for a 4 core Jaguar with a 25w TDP. If high clocks with low TDP was possible, they would have already done it. PS3 is a Cell design which is totally different from the PS4 AMD design. When you drop a process node, with the same overall design, you either gain higher clocks with the same TDP or lower TDP with the same clocks. I don't know how much of a drop it is going from 40nm to 28nm, but if AMD is saying a 10% clock boost with a slight increase in TDP, I don't see how anyone can come to the conclusion that 3Ghz+ is easy. If it was, then they would have done it. It's not like the TDP of the APU is going to be like that of a desktop either since there are power limits to what a console system can pull from the wall.
Posted on Reply
#19
theoneandonlymrk
mastrdrver said:
PS4 is notin and Richland. Richland does not have Jaguar cores. Jaguar is an updated Bobcat core, not Bulldozer.



AMD has already stated what max clocks will be for a 4 core Jaguar with a 25w TDP. If high clocks with low TDP was possible, they would have already done it. PS3 is a Cell design which is totally different from the PS4 AMD design. When you drop a process node, with the same overall design, you either gain higher clocks with the same TDP or lower TDP with the same clocks. I don't know how much of a drop it is going from 40nm to 28nm, but if AMD is saying a 10% clock boost with a slight increase in TDP, I don't see how anyone can come to the conclusion that 3Ghz+ is easy. If it was, then they would have done it. It's not like the TDP of the APU is going to be like that of a desktop either since there are power limits to what a console system can pull from the wall.
Sos kabini I miswaffled. Put kabini In and my statement stands top bin kabini is ps4 and one t two cores could easy do 3ghz in tdp with scaled p states
Posted on Reply
#20
NeoXF
(ignoring core count and IPC gains) E2-2000 1,75GHz @ 18W TDP -> +10% clock = 1,925Ghz @ 18W+ -> 25W ~= 2GHz+ final clocks, for the X4 5110 (25W TDP) flagship. Refreshes will probably go something like 10% beyond that as well (like E-450 to E2-1800 to E2-2000).
Posted on Reply
Add your own comment